1 Introduction

Fluid resuscitation is one of the vital components of hemodynamic management in critically ill patients. To ensure adequate tissue perfusion it should be targeted by optimizing cardiac stroke volume (SV) while limiting increases in filling pressures and development of edema. To achieve this goal, absolute values of SV and cardiac output are less important than monitoring trends in response to interventions such as a fluid challenge [1]. Therefore, continuous monitoring of changes in SV during fluid resuscitation is desirable.

Pulmonary artery catheterization (PAC) is an invasive method that provides continuous assessment of cardiac output in critically ill patients. However, due to the high risk for complications and questionable clinical benefit, PAC use is declining in clinical practice [2]. Transpulmonary thermodilution (TPTD) cardiac output with continuous monitoring by pulse contour analysis is a clinically established, and potentially less invasive, alternative [3]. Even though TPTD with pulse contour analysis is less invasive than PAC, it still requires placement of the central venous and peripheral arterial lines, and provides continuous data with limited reliability, requiring frequent re-calibration [4].

Thoracic electrical impedance tomography (EIT) is a noninvasive imaging modality providing real-time data on changes of electrical bioimpedance [5]. EIT can be used to track changes in pulmonary air content [6, 7] and, when used and interpreted in conjunction with a specific ventilator protocol, recruitment and derecruitment as well as relative hyperdistension [8, 9]. As pulsatile cardiovascular activity generates measurable changes in bioimpedance in the heart and lung regions, it has been presumed that EIT could be able to noninvasively track changes in SV, potentially making it a valuable monitoring tool for patients in need of hemodynamic optimization.

Vonk-Noordegraaf et al. [10] were the first to report the measurement of SV by means of EIT via the cardiosynchronous impedance change in the heart region (\(\Delta Z_{\text{H}}\)) in humans. In two recent studies this was further investigated in pigs. Pikkemaat and colleagues [11] revealed certain issues (sensitivity to heart displacement and lung volume) of the heart-based SV estimation which require further investigation. Even though not reported in [11], the same study also estimated SV via the cardiosynchronous impedance change in the lung region (\(\Delta Z_{\text{L}}\)) with comparable results to the heart-based approach [12]. The approach of tracking SV by analyzing changes in \(\Delta Z_{\text{L}}\) was then further studied by the group of da Silva Ramos [13] who showed that (1) assessment of trending relative SV changes is feasible while (2) estimating absolute SV values requires anthropometric calibration. As also discussed in [14], these partly contradictory findings from animal studies [11,12,13] are not conclusive about the use of cardiosynchronous impedance changes in the heart (\(\Delta Z_{\text{H}}\)) or lung region (\(\Delta Z_{\text{L}}\)) as noninvasive surrogate measures of SV.

As all the recent work on noninvasive assessment of SV via EIT is based on animal experiments, we intended to investigate the applicability of this method in patients in the intensive care unit (ICU) receiving a fluid challenge. In particular we tested the two hypotheses that either \(\Delta Z_{\text{H}}\) or \(\Delta Z_{\text{L}}\) can be used to assess changes in SV by means of EIT.

2 Methods

2.1 Study population, data acquisition and study protocol

Between August 2016 and December 2017, we performed a prospective observational study including critically ill mechanically ventilated patients requiring fluid administration as per clinical decision under monitoring with both EIT and TPTD cardiac output. Exclusion criteria were: age below 18 years, BMI of more than 35 kg/m2, open lung injuries, instable injuries of spine or skull, metallic implants in the thorax region (e.g. cardiac pacemaker). The study was approved by the local Ethics Committee of the Medical Faculty of the Christian-Albrechts University Kiel, Germany (EC-Study-ID D486/16) and registered prior to patient enrollment at ClinicalTrials.gov (NCT02992002, Principal investigator: Tobias Becher, Date of registration: December 14, 2016). Due to the purely observational nature of the study, informed consent was waived by the ethics committee. Some results of this study unrelated to SV analysis have been previously published [15].

All measurements were performed in the ICU facilities of the University Medical Center Schleswig-Holstein (UKSH, Kiel, Germany). The patients were equipped with a 16-electrode EIT belt placed around their chest at the level of the 4th to 5th intercostal space and connected to the PulmoVista™ 500 EIT device (Dräger Medical, Lübeck, Germany). Hemodynamic data including ECG were measured using a Datex-Ohmeda hemodynamic monitor (S/5, Datex-Ohmeda, Helsinki, Finland) and recorded via the S/5 Collect software (Datex-Ohmeda, Helsinki, Finland). SV reference measurements were obtained by averaging three TPTD measurements performed with the PiCCO plus device (Pulsion, Munich, Germany). Figure 1 shows the schematic drawing of the measurement setup.

Fig. 1
figure 1

Block diagram of the measurement setup. For each SV measurement the EIT images are averaged to one representative cardiac cycle (Imgs) from which regions of interest representing the heart (ROIH) and lung regions (ROIL) are extracted. The amplitudes of the temporal sum signals in these two regions are further compared to the SV reference measurement SVREF

Full size image

For the present analysis, SV reference values were obtained from repeated TPTD measurements at three different time points (M1 to M3): (M1) 30 min after application of the EIT belt and prior to the fluid challenge via the injection of 500 mL of balanced electrolyte solution (Sterofundin ISO, B. Braun, Melsungen, Germany); (M2) right after the fluid challenge; (M3) 30 min after (M2). For those patients being fluid responsive, an increase in SV of at least 15% was expected from (M1) to (M2). Moreover, if all of the injected liquid remained in the circulation, SVs of (M3) and (M2) were expected to be comparable.

2.2 Data preprocessing

First, EIT and hemodynamic data were manually synchronized in the time domain with the help of deliberate spikes induced via synchronous tapping on EIT and ECG electrodes as detailed in [14]. All heartbeats in a 2 min window centered around the mean time of injection of the TPTD triplets were averaged via ECG-gated ensemble averaging to one representative cardiac cycle per measurement (M1 to M3) as further described in [14]. EIT raw voltages were reconstructed into sequences of images with 32 × 32 pixels using the GREIT algorithm [16] with the recommended parameters in combination with the 2.5D model of an adult human thorax available in the EIDORS toolbox [17]. Then, for each resulting EIT image sequence, regions of interest were determined as described in [14, 18] for the heart (ROIH) and via the algorithm proposed in [19, 20] for the lungs (ROIL). In brief, these algorithms exploit the amplitude and phase information of the EIT image sequence at the cardiac frequency to separate the heart from the lungs. The cardiosynchronous impedance changes in the heart and the lung regions both exhibit high amplitudes but of opposite phase, which are attributable to blood leaving the heart while the lungs are being filled with blood and vice versa. The resulting two regions of interest were then applied to separately analyze impedance changes in the heart (ROIH) and lung (ROIL) regions.

2.3 Signal processing and data analysis

As mentioned in the introduction, we tested two hypotheses: whether changes in SV can be assessed by means of EIT via the signal amplitude in either (a) the heart [10, 11] or (b) the lung region [12, 13]. To do so, the amplitudes \(\Delta {\text{Z}}_{\text{H}}\) and \(\Delta {\text{Z}}_{\text{L}}\) of the temporal sum signals in both ROIH and ROIL were estimated for each of the averaged EIT image sequences as the peak-to-peak amplitude of the corresponding temporal sum signal, i.e. the sum over all pixels in the corresponding region. To allow for a fair comparison between the three measurements (M1 to M3) of each patient, the abovementioned amplitudes were calculated from the same heart and lung regions which were averaged from the three individual regions of measurements at time points M1 to M3.

The SV trending ability of EIT was tested by transforming the two amplitude features (\(\Delta {\text{Z}}_{\text{H}}\) and \(\Delta {\text{Z}}_{\text{L}}\)) into relative changes between the current measurement \(c\) and a baseline measurement \(b\): \(\Delta {\text{SV}}_{\text{EIT}}^{b \to c} = \left( {\Delta {\text{Z}}_{\text{X}}^{c} {-} \Delta {\text{Z}}_{\text{X}}^{b} } \right)/\Delta {\text{Z}}_{\text{X}}^{b}\), where \(\Delta {\text{Z}}_{\text{X}}\) represents \(\Delta {\text{Z}}_{\text{H}}\) or \(\Delta {\text{Z}}_{\text{L}}\). The reference measurements \({\text{SV}}_{\text{Ref}}\) were transformed similarly and denoted as \(\Delta {\text{SV}}_{\text{Ref}}\).

As further explained in the results and discussion sections, the relative changes of the two transitions M1 to M2 and M2 to M3 were tested for each patient, i.e. \(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M1}} \to {\text{M2}}}}\) versus \(\Delta {\text{SV}}_{\text{EIT}}^{{{\text{M1}} \to {\text{M2}}}}\) and \(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M2}} \to {\text{M3}}}}\) versus \(\Delta {\text{SV}}_{\text{EIT}}^{{{\text{M2}} \to {\text{M3}}}}\) were compared by means of four-quadrant plot analysis [21, 22]. The resulting trending performance was evaluated by means of the concordance rate CR (percentage of data points lying in the 1st and 3rd quadrant, as thoroughly discussed in [21, 22] and Pearson’s correlation coefficient R. According to the criteria defined by Critchley [21] trending performance was considered as acceptable if CR ≥ 92%. Variations of \(\Delta {\text{SV}}_{\text{Ref}}\) and \(\Delta {\text{SV}}_{\text{EIT}}\) below ± 13% were excluded from four-quadrant plot analysis (exclusion zone) as this is the error to be expected from averaging a triplet of TPTD measurements [23, 24].

In a second analysis the relative SV changes were transformed into absolute SV values by using one TPTD measurement as calibration, i.e. \({\text{SV}}_{\text{EIT}}^{\text{M1}}\) and \({\text{SV}}_{\text{EIT}}^{\text{M3}}\) (expressed in mL) were obtained by calibrating \(\Delta {\text{SV}}_{\text{EIT}}^{{{\text{M1}} \to {\text{M2}}}}\) and \(\Delta {\text{SV}}_{\text{EIT}}^{{{\text{M2}} \to {\text{M3}}}}\) against \({\text{SV}}_{\text{Ref}}^{\text{M2}}\). The noninvasive estimates \({\text{SV}}_{\text{EIT}}^{\text{M1}}\) and \({\text{SV}}_{\text{EIT}}^{\text{M3}}\) were then compared to \({\text{SV}}_{\text{Ref}}^{\text{M1}}\) and \({\text{SV}}_{\text{Ref}}^{\text{M3}}\) by means of Bland–Altman analysis.

2.4 Rejection of outliers caused by variations in tidal volume ratio

It is known that changes in tidal volume (VT) correlate well with changes in global EIT impedance [25, 26]. Therefore, the ratio of EIT-derived VT versus true—ventilator-derived—VT should remain constant over time if no perturbation (such as EIT belt displacement or EIT electrode contact problem) occurred. As such perturbations can also impair the EIT-based hemodynamic parameter estimation, potential outliers were identified and removed from the analysis. To this end, for each measurement (M1 to M3), VT was estimated in two ways: (a) from the ventilator data provided to the EIT machine via the MEDIBUS interface (\({\text{VT}}_{\text{V}}\)) and (b) as the sum of all pixels of a functional ventilation image (standard-deviation in the temporal domain) obtained from band-pass filtered (cutoff frequencies: [0.1, 0.5] Hz) EIT images (\({\text{VT}}_{\text{E}}\)). Then the ratio of these VT was calculated as \({\text{VT}}_{\text{E/V}} = {\text{VT}}_{\text{E}} /{\text{VT}}_{\text{V}}\). For each relative measurement M1 to M2 and M2 to M3 the relative change in this tidal volume ratio was calculated as follows: \(\Delta {\text{VT}}_{\text{E/V}}^{{{\text{M1}} \to {\text{M2}}}} = \left( {{\text{VT}}_{\text{E/V}}^{\text{M2}} - {\text{VT}}_{\text{E/V}}^{\text{M1}} } \right)/{\text{VT}}_{\text{E/V}}^{\text{M1}}\) and \(\Delta {\text{VT}}_{\text{E/V}}^{{{\text{M2}} \to {\text{M3}}}} = \left( {{\text{VT}}_{\text{E/V}}^{\text{M3}} - {\text{VT}}_{\text{E/V}}^{\text{M2}} } \right)/{\text{VT}}_{\text{E/V}}^{\text{M2}}\), respectively. The analysis was restricted to measurements which fell into the limits of the 95% confidence interval, i.e. \(\mu_{{\Delta {\text{VT}}_{\text{E/V}} }} - 1.96 \sigma_{{\Delta {\text{VT}}_{\text{E/V}} }} < \Delta {\text{VT}}_{\text{E/V}} < \mu_{{\Delta {\text{VT}}_{\text{E/V}} }} + 1.96 \sigma_{{\Delta {\text{VT}}_{\text{E/V}} }}\), with \(\mu_{{\Delta {\text{VT}}_{\text{E/V}} }}\) and \(\sigma_{{\Delta {\text{VT}}_{\text{E/V}} }}\) as the mean and standard-deviation of all \(\Delta {\text{VT}}_{\text{E/V}}\).

3 Results

In total, 25 patients were included in the study. The average duration of fluid administration was 15 ± 6 (range: 5–28) minutes. 12 out of 25 patients were “fluid responsive”, as defined by an increase in SV of more than 15% after fluid administration.

Since no ECG signal was recorded in the first four patients, only 21 out of 25 patients (15 male/6 female, age: 59 ± 22 years, weight: 78 ± 12 kg, height: 175 ± 7, BMI: 25 ± 3 kg/m2, denoted as P5 to P25) were included in the present analysis. Due to technical issues with the EIT device the last measurement (M3) of subject P21 was not recorded and therefore only \(\Delta {\text{SV}}_{\text{EIT}}^{{{\text{M1}} \to {\text{M2}}}}\) and \(\Delta {\text{SV}}_{\text{SV}}^{{{\text{M1}} \to {\text{M2}}}}\) were compared for this subject. Finally, the measurements of P25 were removed from further analysis since the variations in tidal volume ratios \(\Delta {\text{VT}}_{\text{E/V}}\) exceeded the 95% limits defined by the outlier removal approach. This results in a total of 39 measurements performed in 20 patients remaining for analysis, including 9 fluid “responders” and 11 “non-responders”. Table 1 shows ventilation parameters and SV reference measurements for each patient.

Table 1 Ventilation and hemodynamic parameters for each measurement (M1 to M3) and each patient (P05 to P25)
Full size table

3.1 Hemodynamic variations

The average changes in \({\text{SV}}_{\text{Ref}}\) between the three measurements (M1, M2 and M3) were as follows: the fluid challenge generally led to an increase of SV (M2 vs. M1: \(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M1}} \to {\text{M2}}}}\) = 13.1%) while 30 min after SV returned to its baseline level (M2 vs M3: \(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M2}} \to {\text{M3}}}}\) = − 10.0%). Relative changes in SV and \(\Delta {\text{VT}}_{\text{E/V}}\) and the EIT-derived amplitudes \(\Delta {\text{Z}}_{\text{H}}\) and \(\Delta {\text{Z}}_{\text{L}}\) are listed in Table 2 for each patient.

Table 2 Relative changes of reference stroke volume (ΔSVRef) and EIT-derived parameters between measurements M1 and M2 and M2 and M3, denoted as M1 → M2 and M2 → M3, respectively
Full size table

3.2 SV trending performance of EIT

The trending performance for all 39 measurements and the two EIT-based amplitude features (\(\Delta {\text{Z}}_{\text{H}}\) and \(\Delta {\text{Z}}_{\text{L}}\)) is shown in Fig. 2a and b by means of four-quadrant plot analysis. CR of 52.6% and R of 0.13 (P = 0.44) were obtained for the heart-related amplitude \(\Delta {\text{Z}}_{\text{H}}\) (Fig. 2a). In contrast, using the lung-related amplitude \(\Delta {\text{Z}}_{\text{L}}\) led to CR of 94.4% and R of 0.72 (P < 0.0001) as shown in Fig. 2b.

Fig. 2
figure 2

a and b: Trending analyses between \(\Delta {\text{SV}}_{\text{EIT}}\) versus \(\Delta {\text{SV}}_{\text{Ref}}\) by means of four-quadrant plot analysis for a the heart amplitude \(\Delta {\text{Z}}_{\text{H}}\) and b the lung amplitude \(\Delta {\text{Z}}_{\text{L}}\). The performance is assessed using the concordance rate CR and Pearson’s correlation coefficient R. Each analysis includes a total of 39 measurement points (from 20 patients) representing relative changes from M2 to M3 (symbols with black border) and changes from M1 to M2 (symbols with no border). c and d: Bland–Altman analysis for c heart and d lung amplitude comparing \({\text{SV}}_{\text{EIT}}^{\text{M1}}\) versus \({\text{SV}}_{\text{Ref}}^{\text{M1}}\) (no border) and \({\text{SV}}_{\text{EIT}}^{\text{M3}}\) versus \({\text{SV}}_{\text{Ref}}^{\text{M3}}\) (with black border) after calibration against \({\text{SV}}_{\text{Ref}}^{\text{M2}}\)

Full size image

3.3 Absolute SV measurements after calibration

Assuming a calibration with TPTD at the time of measurement M2, absolute SV values were obtained for measurements M1 and M3. The comparison of invasive measurements \({\text{SV}}_{\text{Ref}}^{\text{M1}}\) and \({\text{SV}}_{\text{Ref}}^{\text{M3}}\) versus noninvasive estimates \({\text{SV}}_{\text{EIT}}^{\text{M1}}\) and \({\text{SV}}_{\text{EIT}}^{\text{M3}}\) is shown in Fig. 2c and d by means of Bland–Altman analysis. For the heart-related amplitude \(\Delta {\text{Z}}_{\text{H}}\) in Fig. 2c a bias of 11.9 mL and 95% limits of agreement (LOA) of − 20.4 to 44.1 mL were obtained. The use of the lung-related amplitude \(\Delta {\text{Z}}_{\text{L}}\) shown in Fig. 2d resulted in a bias of 2.7 mL and LOA of − 12.3 to 17.6 mL.

4 Discussion

In the present study, we assessed EIT measurements performed before, during and after a clinically indicated fluid challenge for their ability to track changes in cardiac SV in critically ill adult patients. In particular, we analyzed the changes in cardiosynchronous impedance amplitudes in the heart and lung region and compared them to reference measurements obtained via TPTD. Analyzing the changes in heart-beat-related impedance amplitudes in the lung region, we found a high concordance rate of 94.4% and a strong correlation (R = 0.72 with P < 0.0001) with the reference measurements. Bland–Altman analysis revealed LOA of − 12.3 to 17.6 mL with minimal bias. In contrast, our analysis of the cardiosynchronous impedance amplitudes in the heart region resulted in a weak correlation (R = 0.13 with P = 0.44) and LOA of − 20.4 to 44.1 mL.

4.1 Hemodynamic variations

On average, SV increased most between M1 and M2 (\(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M1}} \to {\text{M2}}}}\) = 13.1%) and decreased most between M2 and M3 (\(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M2}} \to {\text{M3}}}}\) = − 10.0%). This justifies our choice of analyzing the trending ability from M1 to M2 and from M2 to M3, in order to get the lowest possible amount of measurements falling within the error range of the reference device (i.e. the exclusion zone in the four-quadrant plot). On average, we observed only a slight increase in SV in response to the fluid challenge (\(\Delta {\text{SV}}_{\text{Ref}}^{{{\text{M1}} \to {\text{M2}}}}\) = 13.1%). This is due to the fact that only 9 out of 20 patients were “fluid responsive”, as defined by a change in SV of > 15% after fluid challenge [27]. The subsequent return of SV to baseline could be explained by a redistribution of the crystalloid solution used for injection. A few minutes after the fluid challenge with crystalloids, only roughly 25% of the injected volume remains in the circulation [28], explaining the observed return of SV to its baseline level.

4.2 SV trending performance of EIT

As shown in Fig. 2a, the cardiosynchronous amplitude in the heart region, \(\Delta {\text{Z}}_{\text{H}}\), does not provide a reliable estimate (CR = 52.6% and R = 0.13 with P = 0.44) for the noninvasive trending of SV via EIT. In contrast, the cardiosynchronous amplitude in the lung region, \(\Delta {\text{Z}}_{\text{L}}\), shown in Fig. 2b indicates an acceptable trending performance with CR = 94.4% and R = 0.72 (P < 0.0001).

Based on the present analysis we must reject the hypothesis that the EIT amplitude \(\Delta {\text{Z}}_{\text{H}}\) can be used for trending—i.e. following relative changes—of SV, which is in contradiction with previous studies [10, 11]. We hypothesize that the heart-region-based SV estimation is strongly affected by the placement of the EIT belt with respect to the patient’s heart or by ventilation-induced displacement of the heart. While in the animal experiments by Pikkemaat et al. [11] more accurate belt placements might have led to a less pronounced influence of the first effect in some pigs, it was still mentioned as a main limitation for other pigs [11].

In contrast, our analysis confirms that the EIT amplitude \(\Delta {\text{Z}}_{\text{L}}\) can be successfully used for noninvasive trending of SV. However, this is only possible after exclusion of two measurements with high variations in tidal volume ratios VTE/V. Without this exclusion a less good performance of CR = 85.0% and R = 0.54 (P < 0.001) was obtained (results not shown). This highlights the sensitivity of our approach to perturbations such as electrode–skin contact problems [29], EIT belt displacement [18] or large changes in ventilation [30]. Thus, contrary to the amplitude in the heart region, SV trending via the amplitude in the lung region might be more promising. This would confirm the recent work by da Silva Ramos and colleagues [13] but is at the same time partly in contradiction with the mixed results obtained by Pikkemaat [12]. Besides, these two studies [12, 13] were performed on pigs under laboratory conditions while our study is based on clinical data of ICU patients and thus reveals more challenges encountered under real-life conditions in a less controlled environment.

4.3 Limitations and future work

The present study is limited in various aspects. First, we could only analyze a total of 39 measurements from 20 patients with rather low variations in SV (\(\left| {\Delta {\text{SV}}_{\text{Ref}} } \right| = 11.7 \pm 11.0\)%, range 0.1% to 43.3%). Even though a passive leg raising maneuver—instead of a fluid challenge—could lead to higher variations in SV, this is not common practice in this ICU due to practical considerations. As the current patient population contains many “non-responders” to fluid, enrolling more patients could increase the \(\Delta {\text{SV}}\) and might also allow to only include those patients being fluid-responsive.

Second, our SV trending approach relies on the cardiosynchronous EIT amplitude in the lung region, which might suggest that \(\Delta {\text{Z}}_{\text{L}}\) represents global perfusion, which is also a widespread assumption [31]. However, the cardiosynchronous EIT signals in the lung region do not necessarily precisely reflect pulmonary perfusion as discussed by Hellige and Hahn [32] and Adler et al. [33]. For example, two branches of pulmonary arteries with equal perfusion but different compliance would result in different cardiosynchronous EIT signal amplitudes [20]. Therefore, it is important to underline that our approach is not intended to detect regional pulmonary perfusion but rather investigates trends in cardiosynchronous impedance changes, as detected in the pulmonary region, to follow changes in cardiac SV. As shown in our Bland–Altman analysis, it can be calibrated using actual SV measurements. This procedure introduces a calibration factor that corrects for inter-individual variability of parameters like pulmonary artery compliance and therefore allows relatively accurate trending of actual SV. Of note, some patients were under high doses of vasopressors which may have an influence on pulmonary vascular resistance and may decrease hypoxic pulmonary shunt. Since we did not measure pulmonary arterial pressure in our study, we were not able to assess the actual impact of pulmonary vascular resistance on our findings. Nevertheless, it appears that our calibration procedure led to reliable estimations of changes in SV even in patients under high doses of vasopressors.

Even though the present approach shows promise on the current data, further research—including long-term measurements and patients with various pathophysiological conditions—is required to reveal possible limitations of this approach. Furthermore, the use of hypertonic saline injection as EIT contrast agent [34,35,36] could help to reveal potential discrepancies between the cardiosynchronous EIT signal in the lung region and the real pulmonary perfusion.

Third, since all patients were on pressure-controlled ventilation (PCV) or pressure support ventilation (PSV) we did not assess plateau pressure or static respiratory system compliance. Approximately one-third of patients had relevant spontaneous breathing activity, which makes calculation of compliance during PCV or PSV unreliable [37].

Fourth, the use of ECG-gated EIT averaging is not ideal in patients with cardiac arrhythmia. A more sophisticated method for isolating cardiosynchronous changes form the EIT images will be necessary in this group of patients.

Finally, it has also been shown in healthy volunteers that EIT-based SV estimation is sensitive to perturbations from electrode–skin contact problems or changes in posture [29]. The latter has also been reported for cardiac EIT signals in general [30, 38]. This might limit the present approach to scenarios where minimal changes in the EIT measurement setup occur (i.e. stable ventilation, stable electrode contact and unchanged posture).

For future studies, it might be of interest to investigate the trending capability in a patient population with higher intra-subject variations of SV (e.g. during hemorrhage or fluid resuscitation) and to assess the effects of different fluids (crystalloid, colloid, blood, etc.). Moreover, a specific protocol including different ventilation maneuvers would further allow to better identify the limitations of our approach related to changes in lung volume. Finally, the long-term trending performance of our noninvasive approach could also be compared to continuous pulse contour analysis with intermittent TPTD measurements.

5 Conclusion

In our study, we demonstrated the ability of EIT to assess changes in SV by analyzing cardiosynchronous impedance changes in the lung region. This approach is limited by the influences induced by ventilation, posture or electrode contact impedance. Future work should thus investigate potential limitations of this approach in detail. If it shows acceptable in certain situations, EIT could be a useful tool for noninvasive monitoring of SV trends at the bedside.